WELFARE ECONOMICS

(DRAFT, SEPTEMBER 22, 2006)

W000031

In 1776, the same year as the American Declaration of Independence, Adam Smith
published The Wealth of Nations. Smith laid out an argument that is now familiar to all
economics students: (1) The principal human motive is self-interest. (2) The invisible
hand of competition automatically transforms the self-interest of many into the common
good. (3) Therefore, the best government policy for the growth of a nation’s wealth is
that policy which governs least.
Smith’s arguments were at the time directed against the mercantilists, who
promoted active government intervention in the economy, particularly in regard to (ill-
conceived) trade policies. Since his time, his arguments have been used and reused by
proponents of laissez-faire throughout the 19th and 20th centuries. Arguments of Smith
and his opponents are still very much alive today: The pro-Smithians are those who place
their faith in the market, who maintain that the provision of goods and services in society
ought to be done, by and large, by private buyers and sellers acting in competition with
each other. One can see the spirit of Adam Smith in economic policies involving
deregulation, tax reduction, denationalizing industries, and reduction in government
growth in western countries; and in the deliberate restoration of private markets in China,
the former Soviet Union, and other eastern European countries. The anti-Smithians are
also still alive and well; mercantilists are now called industrial policy advocates, and
there are intellectuals and policy makers who believe that: (1) economic planning is
superior to laissez-faire; (2) markets are often monopolized in the absence of government
intervention, crippling the invisible hand of competition; (3) even if markets are
competitive, the existence of external effects, public goods, information asymmetries and
other market failures ensure that laissez-faire will not bring about the common good; (4)
and in any case, laissez-faire may produce an intolerable degree of inequality.
The branch of economics called welfare economics is an outgrowth of the
fundamental debate that can be traced back to Adam Smith, if not before. It is the
economic theory of measuring and promoting social welfare.
This entry is largely organized around three propositions. The first answers this
question: In an economy with competitive buyers and sellers, will the outcome be for the
common good? The second addresses the issue of distributional equity, and answers this
question: In an economy where distributional decisions are made by an enlightened
sovereign, can the common good be achieved by a slightly modified market mechanism,
or must the market be abandoned? The third focuses on the general issue of defining
social welfare, or the common good, whether via the market, via a centralized political
process, or via a voting process. It answers this question: Does there exist a reliable way
to derive the true interests of society, regarding, for example, alternative distributions of
income or wealth, from the preferences of individuals?
This entry focuses on theoretical welfare economics. There are related topics in
practical welfare economics which are only mentioned here. A reader interested in the
practical problems of evaluating policy alternatives can refer to entries on
CONSUMERS’ SURPLUS, COST-BENEFIT ANALYSIS and COMPENSATION
PRINCIPLE, to name a few.

I. The First Fundamental Theorem, or Laissez-Faire Leads to the

Common Good
‘The greatest meliorator of the world is selfish, huckstering trade.’ (R.W.
Emerson, Work and Days)
In The Wealth of Nations, Book IV, Smith wrote: ‘Every individual necessarily labours to
render the annual revenue of the society as great as he can. He generally indeed neither
intends to promote the public interest, nor knows how much he is promoting it … . He
intends only his own gain, and he is in this, as in many other cases, led by an invisible
hand to promote an end which was no part of his intention.’ The First Fundamental
Theorem of Welfare Economics can be traced back to these words of Smith. Like much
of modern economic theory, the First Theorem is set in the context of a Walrasian general
equilibrium model, developed almost a hundred years after The Wealth of Nations. Since
Smith wrote long before the modern mathematical language of economics was invented,
he never rigorously stated, let alone proved, any version of the First Theorem. That was
first done by Lerner (1934), Lange (1942) and Arrow (1951).
To establish the First Theorem, we need to sketch a general equilibrium model of
an economy. Assume all individuals and firms in the economy are price takers: none is
big enough, or motivated enough, to act like a monopolist. Assume each individual
chooses his consumption bundle to maximize his utility subject to his budget constraint.
Assume each firm chooses its production vector, or input–output vector, to maximize its
profits subject to some production constraint. Note that we assume self-interest, or the
absence of externalities: An individual cares only about his own utility, which depends
only on his own consumption. A firm cares only about its own profits, which depend only
on its own production vector.
The invisible hand of competition acts through prices; they contain the information
about desire and scarcity that coordinate actions of self-interested agents. In the general
equilibrium model, prices adjust to bring about equilibrium in the market for each and
every good. That is, prices adjust until supply equals demand. When that has occurred,
and all individuals and firms are maximizing utilities and profits, respectively, we have a
competitive equilibrium.
The First Theorem establishes that a competitive equilibrium is for the common
good. But how is the common good defined? The traditional definition looks to a
measure of total value of goods and services produced in the economy. In Smith, the
‘annual revenue of the society’ is maximized. In Pigou (1920), following Smith, the ‘free
play of self-interest’ leads to the greatest ‘national dividend’.
However, the modern interpretation of ‘common good’ typically involves Pareto
optimality, rather than maximized gross national product. When ultimate consumers
appear in the model, a situation is said to be Pareto optimal if there is no feasible
alternative that makes everyone better off. Pareto optimality is thus a dominance concept
based on comparisons of vectors of utilities. It rejects the notion that utilities of different
individuals can be compared, or that utilities of different individuals can be summed up
and two alternative situations compared by looking at summed utilities. When ultimate
consumers do not appear in the model, as in the pure production framework to be
described below, a situation is said to be Pareto optimal if there is no alternative that
results in the production of more of some output, or the use of less of some input, all else
equal. Obviously saying that a situation is Pareto optimal is not the same as saying it
maximizes GNP, or that it is best in some unique sense. There are generally many Pareto
optima. However, optimality is a common good concept that can get common assent: No
one would argue that society should settle for a situation that is not optimal, because if A
is not optimal, there exists a B that all prefer.
In spite of the multiplicity of optima in a general equilibrium model, most states are
non-optimal. If the economy were a dart board and consumption and production
decisions were made by throwing darts, the chance of hitting an optimum would be zero.
Therefore, to say that the market mechanism leads an economy to an optimal outcome is
to say a lot. And now we can turn to a modern formulation of the First Theorem:
First Fundamental Theorem of Welfare Economics: Assume that all individuals
and firms are self-interested price takers. Then a competitive equilibrium is Pareto
optimal.
To illustrate the theorem, we focus on one simple version of it, set in a pure production
economy. For a general versions of the theorem, with both production and exchange, the
reader can refer to Mas-Colell, Whinston & Green (1995).
In a general equilibrium production economy model, there are K firms and m
goods, but, for simplicity, no consumers. We write k = 1, 2,..., K for the firms, and
j = 1, 2,..., m for the goods. Given a list of market prices, each firm chooses a feasible
input–output vector yk so as to maximize its profits. We adopt the usual sign convention
for a firm’s input–output vector yk: ykj < 0 means firm k is a net user of good j, and ykj > 0
means firm k is a net producer of good j. When we add the amounts of good j over all the
firms, y1j + y2j + … + yKj , we get the aggregate net amount of good j produced in the
economy, if positive, and an aggregate net amount of good j used, if negative. What is
feasible for firm k is defined by some fixed production possibility set Yk. Under the sign
convention on the input–output vector, if p is a vector of prices, firm k’s profits are given
by
π k = p ⋅ yk .
A list of feasible input–output vectors y = (y1, y2,…yK) is called a production plan for the
economy. A competitive equilibrium is a production plan ŷ and a price vector p such
that, for every k , yˆ k maximizes πk subject to yk’s being feasible. (Since the production
model abstracts from the ultimate consumers of outputs and providers of inputs, the
supply equals demand requirement for an equilibrium is moot).
If y = (y1, y2, …, yK) and z = (z1, z2, …, zK) are alternative production plans for the
economy, z is said to dominate y if the following vector inequality holds:

k
∑ z k ≥ ∑ yk .
k
The production plan y is said to be Pareto optimal if there is other production plan that
dominates it. (Note that for two vectors a and b, a ≥ b means a j ≥ b j for every good j,
with the strict inequality holding for at least one good.)
We now have the apparatus to state and prove the First Theorem in the context of
the pure production model:
First Fundamental Theorem of Welfare Economics, Production Version. Assume
that all prices are positive, and that ŷ , p is a competitive equilibrium. Then ŷ is
Pareto optimal.
To see why, suppose to the contrary that a competitive equilibrium production plan
yˆ = ( yˆ1 , yˆ 2 ,…, yˆ K ) is not optimal. Then there exists a production plan z = ( z1 , z2 ,…, z K )
that dominates it. Therefore

k
∑ zk ≥ ∑ yˆ k .
k
Taking the dot product of both sides with the positive price vector p gives
p ⋅ ∑ zk > p ⋅ ∑ yˆ k .
k k
But this implies that, for at least one firm k,
p ⋅ zk > p ⋅ yˆ k ,
which contradicts the assumption that yˆ k maximizes firm k’s profits. Q.E.D.

II. First Fundamental Theorem Drawbacks, and the Second

Fundamental Theorem
The First Theorem of Welfare Economics is mathematically true but nevertheless
open to objections. Here are the commonest: (1) The theorem is an abstraction that
ignores the facts. Preferences of consumers are not given, they are created by advertising.
The real economy is never in equilibrium, most markets are characterized by excess
supply or excess demand, and are in a constant state of flux. The economy is dynamic,
tastes and technology are constantly changing, whereas the model assumes they are fixed.
The cast of characters in the real economy is constantly changing, the model assumes it
fixed. (2) The theorem assumes competitive behaviour, whereas the real world is full of
monopoly and market power. (3) The theorem assumes there are no externalities. In fact,
if in an exchange economy person l’s utility depends on person 2’s consumption as well
as his own, the theorem does not hold. Similarly, if in a production economy firm k’s
production possibility set depends on the production vector of some other firm, the
theorem breaks down. In a similar vein, the theorem assumes there are no public goods,
that is, goods like national defense, judicial systems, or lighthouses, that are necessarily
non-exclusive in use. If such goods are privately provided (as they would be in a
completely laissez-faire economy), then their level of production will be sub-optimal. (4)
The theorem ignores distribution. Laissez-faire may produce a Pareto optimal outcome,
but there are many different Pareto optima, and some are fairer than others. Some people
are endowed with resources that make them extremely rich, while others, through no fault
of their own, are extremely poor.
The first and second objections to the First Theorem are beyond the scope of this
entry. The third, regarding externalities and public goods, is one that economists have
always acknowledged. The standard remedies for these market failures involve various
modifications of the market mechanism, including Pigovian taxes (Pigou, 1920) on
harmful externalities, or appropriate Coasian legal entitlements to, for example, clean air
(Coase, 1960).
The important contribution of Pigou is set in a partial equilibrium framework, in
which the costs and benefits of a negative externality can be measured in money terms.
Suppose that a factory produces gadgets to sell at some market-determined price, and
suppose that, as part of its production process, the factory emits smoke which damages
another factory located downwind. In order to maximize its profits, the upwind factory
will expand its output until its marginal cost equals price. But each additional gadget it
produces causes harm to the downwind factory – the marginal external cost of its activity.
If the factory manager ignores that marginal external cost, he will create a situation that is
non-optimal in the sense that the aggregate net value of both firms’ production decisions
will not be as great as it could be. That is, what Pigou calls ‘social net product’ will not
be maximized, although ‘trade net product’ for the polluting firm will be. Pigou’s remedy
was for the state to eliminate the divergence between trade and social net product by
imposing appropriate taxes (or, in the case of beneficial externalities, bounties). The
Pigovian tax would be set equal to marginal external cost, and with it in place the gap
between the polluting firm’s view of cost and society’s view would be closed. Optimality
would be re-established.
Coase’s contribution was to emphasize the reciprocal nature of externalities and to
suggest remedies based on common law doctrines. In his view the polluter damages the
pollutee only because of their proximity, e.g., the smoking factory harms the other only if
it happens to locate close downwind. Coase rejects the notion that the state must step in
and tax the polluter. The common law of nuisance can be used instead. If the law
provides a clear right for the upwind factory to emit smoke, the downwind factory can
contract with the upwind factory to reduce its output, and if there are no impediments to
bargaining, the two firms acting together will negotiate an optimal outcome.
Alternatively, if the law establishes a clear right for the downwind factory to recover for
smoke damages, it will collect external costs from the polluter, and thereby motivate the
polluter to reduce its output to the optimal level. In short, a legal system that grants clear
rights to the air to either the polluter or pollutee will set the stage for an optimal outcome,
provided that bargaining is costless. If bargaining is costly, then the law should be
designed with an eye towards minimizing social costs created by the externality.
With respect to public goods, since Samuelson (1954) derived formal optimality
conditions for their provision, the issue has received much attention from economists;
one especially notable theoretical question has to do with discovering the strengths of
people’s preferences for a public good. If the government supplies a public judicial
system, for instance, how much should it spend on it (and tax for it)? At least since
Samuelson, it has been known that financing schemes like those proposed by Lindahl
(1919), where an individual’s tax is set equal to his marginal benefit, provide perverse
incentives for people to misrepresent their preferences. Schemes that are immune to such
misrepresentations (in certain circumstances) have been developed (Clarke, 1971; Groves
and Loeb, 1975).
But it is the fourth objection to the First Theorem that may be most fundamental.
What about distribution?
 There are two polar approaches to rectifying the distributional inequities of laissez-
faire. The first is the command economy approach: a central bureaucracy makes detailed
decisions about the consumption decisions of all individuals and production decisions of
all producers. The main theoretical problem with the command approach is that it fails to
create appropriate incentives for individuals and firms. On the empirical side, the
experience of the late Soviet and Maoist command economies establish that highly
centralized economic decision making leaves much to be desired, to put it mildly.
The second polar approach to solving distribution problems is to transfer income or
purchasing power among individuals, and then to let the market work. The only kind of
purchasing power transfer that does not cause incentive-related losses is the lump-sum
money transfer. Enter at this point the standard remedy for distribution problems, as put
forward by market-oriented economists, and our second major theorem.
The Second Fundamental Theorem of Welfare Economics establishes that the
market mechanism, modified by the addition of lump-sum transfers, can achieve virtually
any desired optimal distribution. Under more stringent conditions than are necessary for
the First Theorem, including assumptions regarding quasi-concavity of utility functions
and convexity of production possibility sets, the Second Theorem gives the following:
Second Fundamental Theorem of Welfare Economics. Assume that all individuals
and producers are self-interested price takers. Then almost any Pareto optimal
equilibrium can be achieved via the competitive mechanism, provided appropriate
lump-sum taxes and transfers are imposed on individuals and firms.
One version of the Second Theorem, restricted to a pure production economy, is
particularly relevant to an old debate about the feasibility of socialism, see particularly
Lange and Taylor (1939) and Lerner (1944). Anti-socialists including Von Mises (1937)
argued that informational problems would make it impossible to coordinate production in
a socialist economy; while pro-socialists, particularly Lange, argued that those problems
could be overcome by a central planning board, which limited its role to merely
announcing a price vector. This was called ‘decentralized socialism’. Given the prices,
managers of production units would act like their capitalist counterparts; in essence, they
would maximize profits. By choosing the price vectors appropriately, the central planning
board could achieve any optimal production plan it wished.
In terms of the production model given above, the production version of the Second
Theorem is as follows:
Second Fundamental Theorem of Welfare Economics, Production Version. Let ŷ
be any optimal production plan for the economy. Then there exists a price vector
p such that ŷ , p is a competitive equilibrium. That is, for every k, yˆ k maximizes
π k = p ⋅ yk subject to yk being feasible.
The proof of the Second Theorem will not be presented here.

III. Adjusting the Economy and Voting

We rarely choose between a laissez-faire economy and a command economy. Our
choices are almost always more modest. When choosing among alternative tax policies,
or trade and tariff policies, or development policies, or antimonopoly policies, or labour
policies, or transfer policies, what shall guide the choice? The applied welfare
economist’s advice is usually based on some notion of increasing total output in the
economy. The practical political decision, in a democracy, is normally based on voting.

Applied Welfare Economics

The applied welfare economist usually focuses on ways to increase total output,
‘the size of the pie’, or at least to measure changes in the size of the pie. Unfortunately,
theory suggests that the pie cannot be easily measured. This is so for a number of reasons.
To start, any measure of total output is a scalar, that is, a single number. If the number is
found by adding up utility levels for different individuals, illegitimate interpersonal utility
comparisons are being made. If the number is found by adding up the values of aggregate
net outputs of all goods, there is an index number problem. The value of a production
plan will depend on the price vector at which it is evaluated. But in a general equilibrium
context, the price vector will depend on the aggregate net output vector, which will in
turn depend on the distribution of ownership or wealth among individuals.
An early and crucial contribution to the analysis of whether or not the economic pie
has increased in size was made by Kaldor (1939), who argued that the repeal of the Corn
Laws in England could be justified on the grounds that the winners might in theory
compensate the losers: ‘it is quite sufficient [for the economist] to show that even if all
those who suffer as a result are fully compensated for their loss, the rest of the
community will still be better off than before’. Unfortunately, Scitovsky (1941) quickly
pointed out that Kaldor’s compensation criterion (as well as one proposed around the
same time by Hicks) was inconsistent: Consider a move from situation A to situation B. It
is possible to judge B Kaldor superior to A (the move is an improvement) and
simultaneously judge A Kaldor superior to B (the move back would also be an
improvement). This Scitovsky paradox can be avoided via a two-edged compensation
test, according to which B is judged better than A if (1) the potential gainers in the move
from A to B could compensate the potential losers, and still remain better off, and (2) the
potential losers could not bribe the gainers to forego the move.
However, while Scitovsky’s two-edged criterion has some logical appeal, it still has
a major drawback: it ignores distribution. Therefore, it can make no judgement about
alternative distributions of the same size pie. Even worse, both the Kaldor and the
Scitovsky criteria would approve of a change that makes the wealthiest man in society
richer by $1 billion, while making each of the million poorest people worse off by $999.
This is an judgment that many people would reject as wrong or immoral.
Another important tool for measuring changes in economic welfare is the concept
of consumer’s surplus, which Marshall (1920) defined as the difference between what an
individual would be willing to pay for an object, at most, and what he actually does pay.
With a little faith, the economic analyst can measure aggregate consumers’ surplus (note
the new position of the apostrophe), by calculating an area under a demand curve, and
this is in fact commonly done in order to evaluate changes in economic policy. The
applied welfare economist attempts to judge whether the pie would grow in a move from
A to B by examining the change in consumers’ surplus (plus profits, if they enter the
analysis). Some faith is required because consumers’ surplus, like the Kaldor criterion, is
theoretically inconsistent; see for example Boadway (1974).
Under certain circumstances, however, consumers’ surplus inconsistencies can be
ruled out. In particular, if individual utility functions are all quaisilinear, of the form
ui ( xi ) = vi ( xij , j ≠ m ) + xim , then consumers’ surplus paradoxes disappear. (Here ui ( xi ) is
person i’s utility, as a function of his consumption bundle xi = ( xi1 , xi 2 ,..., xim ) , and the
utility function can be separated into two parts, the first one of which is a function vi (⋅)
which depends on quantities of all the goods except the mth, and the second of which is
simply the quantity of the mth good. The mth good can be interpreted as the “money”
good; all the individuals like it, and value it the same way, with the same marginal
utility, of one.) The assumption of quasilinear preferences is a very strong one if we think
about “real” commodities like wine and bread, but it has a certain intuitive appeal if we
are inclined to believe in utility from “money.”
To sum up this section, although the tools of applied welfare economics are widely
used and very important in practice, in theory they should be viewed with some
skepticism.

Voting
In many cases, interesting decisions about economic policies are made either by
government bureaucracies that are controlled by legislative bodies, or by legislative
bodies themselves, or by elected executives. In short, either directly or indirectly, by
voting. The Second Theorem itself raises questions about distribution that many would
view as essentially political: How should society choose the Pareto-optimal allocation of
goods that is to be reached via the modified competitive mechanism? How should the
distribution of income be chosen? How can the best distribution of income be chosen
from among many Pareto optimal ones? Majority rule is a commonly used method of
choice in a democracy, both for political choices and economic ones, and we now turn
our attention to it.
The practical objections to voting, the fraud, the deception, the accidents of
weather, are well known. To quote Boss Tweed, the infamous 19th century chief of New
York’s Tammany Hall: ‘As long as I count the votes, what are you going to do about it?’
We will examine the theoretical problems.
The central theoretical problem with majority voting has been known since the time
of Condorcet’s Essai sur l’application de l’analyse à la probabilité des décisions rendues
à la pluralité des voix, published in 1785: Voting may be logically inconsistent. The now
standard Condorcet voting paradox assumes three individuals 1, 2 and 3, and three
alternatives x, y and z, where the three voters have the following preferences:
1: x y z
2: y z x
3: z x y
(Following an individual’s number the alternatives are listed in his order of preference,
from left to right.) Majority voting between pairs of alternatives will reveal that x beats y,
y beats z, and, paradoxically, z beats x.
It is now clear that such voting cycles are not peculiar; they are generic, particularly
when the alternatives have a spatial aspect with two or more dimensions (Plott, 1967;
Kramer, 1973.) This can be illustrated by taking the alternatives to be different
distributions of one economic pie. Suppose, in other words, that the distributional issues
raised by the First and Second Theorems are to be ‘solved’ by
majority voting, and assume for simplicity that what is to be divided is a fixed total of
wealth, say 100 units.
Now let x be 50 units for person 1, 30 units for person 2 and 20 units for person 3.
That is, let x = (50, 20, 30). Similarly, let y = (30 50, 20) and z = (20, 30, 50). The result
is that our three individuals have precisely the voting paradox preferences. Nor is this
result contrived, it turns out that all the distributions of 100 units of wealth are connected
by endless voting cycles (see McKelvey, 1976). The reader can easily confirm that for
any distributions u and v, that he may choose, there exists a voting sequence from u to v,
and another back from v to u!
The reality of voting cycles should give pause to the economist who recommends
legislation about economic choices, especially choices among alternative distributions of
income or wealth.

IV. Social Welfare and the Third Fundamental Theorem

How then can economic choices be made; how, for example, might the distribution
problem be solved? One potential answer is to assert the existence of a Bergson (1938)
Economic Welfare Function E(·), that depends on the amounts of non-labour factors of
production employed by each producing unit, the amounts of labour supplied by each
individual, and the amounts of produced goods consumed by each individual. Then solve
the problem by maximizing E(·). If necessary conditions for Pareto optimality are derived
that must hold for any E(·), this exercise is harmless enough; but if a particular E(·) is
assumed and distributional implications are derived from it, then an objection can be
raised: Why that Bergson function E(·), and not a different one?
At first, in his modestly titled “A Difficulty in the Concept of Social Welfare”
(1950), and later, in his classic monograph Social Choice and Individual Values (1963),
Kenneth Arrow brought together both the economic and political streams of thought
sketched above. Arrow’s theorem can be viewed in several ways: it is a statement about
the distributional questions raised by the First and Second Theorems; it is an extension of
the Condorcet voting paradox; it is a statement about the logic of voting; and it is a
statement about the logic of Bergson welfare functions, compensation tests, consumers’
surplus tests, and indeed all the tools of the applied welfare economist. Because of its
central importance, Arrow’s theorem can be justifiably called the Third Fundamental
Theorem of Welfare Economics.
Arrow’s analysis is at a high level of abstraction, and requires some additional
model building. From this point onward we assume a given set of at least 3 distinct
alternatives, which might be allocations in an exchange economy, distributions of wealth,
tax bills in a legislature, or candidates in an election. The alternatives are written x, y, z,
etc. We assume a fixed society of individuals, numbered 1, 2,…, n. Let Ri represent the
preference relation of individual i, so xRiy means person i likes x as well as or better than
y. (Strict preference is shown with a Pi , and indifference with a I i .) A preference profile
for society is a specification of preferences for each and every individual, or
symbolically, R = ( R1 , R2 ,..., Rn ) . We write RS for society’s preference relation, arrived
at in a way yet to be specified.
 Arrow was concerned with the logic of how individual preferences are transformed
into social preferences. That is, how is RS found? Symbolically we can represent the
transformation this way:
R1 , R2 ,…, Rn → RS .
Now if society is to make decisions regarding things like distributions of wealth, it must
‘know’ when one alternative is as good as or better than another, even if both are Pareto
optimal. To ensure it can make such decisions, Arrow requires that RS be complete. That
is, for any alternatives x and y, either xRS y or yRS x (or both, if society is indifferent
between the two). If society is to avoid the illogic of voting cycles, its preferences ought
to be transitive. That is, for any alternatives x, y and z, if xRS y and yRS z , then xRS z .
Following Sen (1970), we call a transformation of preference profiles into complete and
transitive social preference relations an Arrow social welfare function, or more briefly, an
Arrow SWF.
Anyone can make up an Arrow SWF, just as anyone can make up a Bergson
function, or for that matter a simple moral judgment about when one distribution of
wealth is better than another. But arbitrary judgments are unsatisfactory and so are
arbitrary Arrow functions. Therefore, Arrow imposed some reasonable conditions on his
function. Following Sen’s (1970) version of Arrow’s theorem, there are four conditions:
(1) Universality. The function should always work, no matter what individual preferences
might be. It would not be satisfactory, for example, to require unanimous agreement
among all the individuals before determining social preferences. (2) Pareto consistency.
If everyone prefers x to y, then the social preference ought to be x over y. (3)
Independence. Suppose there are two alternative preference profiles for individuals in
society, but suppose individual preferences regarding x and y are exactly the same under
the two alternatives. Then the social preference regarding x and y must be exactly the
same under the two alternatives. In particular, if individuals change their minds about a
third ‘irrelevant’ alternative, this should not affect the social preference regarding x and y.
(4) Non-dictatorship. There should not be a dictator. In Arrow’s abstract model, person i
is a dictator if society always prefers exactly what he prefers, that is, if the Arrow
function transforms Ri directly into RS .
An economist or policy maker who wants an ultimate answer to questions
involving distributions, or questions involving choices among alternatives that are not
comparable under the Pareto criterion, could use an Arrow SWF for guidance.
Unfortunately, Arrow showed that imposing conditions 1 to 4 guarantees that Arrow
functions do not exist:
Third Fundamental Theorem of Welfare Economics. There is no Arrow social
welfare function that satisfies the conditions of universality, Pareto consistency,
independence, non-dictatorship.
In order to illustrate the logic of the theorem, we will use a somewhat stronger
assumption than independence. This assumption is called NIM, or neutrality–
independence–monotonicity, defined as follows: Let V be a group of individuals. Suppose
for some preference profile and some particular pair of alternatives x and y, all members
of V prefer x to y, all individuals not in V prefer y to x, and the social preference is x over
y. Then for any preference profile and any pair of alternatives w and z, if all people in V
prefer w to z, the social preference must be w over z. In short, if V gets its way in one
instance, when everyone opposes it, then it must have the power to do it again, when the
opposition may be weaker.
A group of individuals V is said to be decisive if for all alternatives x and y,
whenever all the people in V prefer x to y, society prefers x to y. Assumption NIM asserts
that if V prevails when it is opposed by everyone else, it must be decisive. If the social
choice procedure is majority rule, for example, any group of (n+1)/2 members, for n odd,
or (n/2)+1 members, for n even, is decisive. Moreover, it is clear that majority rule
satisfies the NIM assumption, since if V prevails for a particular x and y when everyone
outside of V prefers y to x, then V must be a majority, and must always prevail. (Majority
rule is just one example of a procedure that satisfies NIM; there are many other
procedures that also do so.)
Now we are ready to turn to a short and simple version of the Third Theorem.
Third Fundamental Theorem of Welfare Economics, Short Version. There is no
Arrow SWF that satisfies the conditions of universality, Pareto consistency,
neutrality–independence–monotonicity, and non-dictatorship.
The proof goes as follows: First, there must exist decisive groups of individuals,
since by the Pareto consistency requirement the set of all individuals is one. Now let V be
a decisive group of minimal size. If there is just one person in V, he is a dictator. Suppose
then that V includes more than one person. We show this leads to a contradiction.
If there are two or more people in V, we can divide it into non-empty subsets V1 and
V2. Let V3 represent all the people who are in neither V1 nor V2. (V3 may be empty). By
universality, the Arrow function must be applicable to any profile of individual
preferences. Take three alternatives x, y and z and consider the following preferences
regarding them:
For individuals in V1 : x y z
For individuals in V2 : y z x
For individuals in V3 : z x y
(At this point the close tie between Arrow and Condorcet is clear, for these are exactly
the voting paradox preferences!)
Since V is by assumption decisive, y must be socially preferred to z, which we
write yPS z . By the assumption of completeness for the social preference relation, either
xRS y or yPS x must hold. If xRS y holds, since xRS y and yPS z , then xPS z must hold by
transitivity. But now V1 is decisive by the NIM assumption, contradicting V’s minimality.
Alternatively, if yPS x holds, V2 is decisive by the NIM assumption, again contradicting
V’s minimality. In either case, the assumption that V has two or more people leads to a
contradiction. Therefore V must contain just one person, who is, of course, a dictator!
Q.E.D.
Since the Third Theorem was discovered, a whole literature of modifications and
variations has been spawned. But the depressing conclusion has remained more or less
the same: there is no logically infallible way to aggregate the preferences of diverse
individuals into a social preference relation. Therefore, there are no logically infallible
ways to vote, or to solve the problems of distribution of income and wealth in society.
V. Social Welfare After Arrow

Social Choice Functions and Strategy

The Third Fundamental Theorem of Welfare Economics tells us that we cannot find an
Arrow social welfare function satisfying certain reasonable requirements. An Arrow
function maps preference profiles (that is, preference relations for each and every
member of society) into social preference relations. But in order to make judgments
about what alternative is best for society, it is not really necessary to have a social
preference relation. Suppose we just had a rule that tells us, if the set of alternatives is x,
y, z, etc., and the preference profile is R = ( R1 , R2 ,..., Rn ) , then the best alternative is
such-and-such? Such a rule would be a mapping from preference profiles into
alternatives, written symbolically as follows:
R1 , R2 ,…, Rn → x.
Such a rule is called a social choice function, or SCF for short. An Arrow function
produces a social ranking of all the alternatives; an SCF in contrast, just produces a
winner. As an example, think of plurality voting, with some kind of rule to break ties.
The essential difficulty with SCF’s is that they may create obvious incentives for
people to misrepresent their preferences, so as to obtain better (for them) social choices.
As an example, consider again the Condorcet voting paradox preferences:
1: x y z
2: y z x
3: z x y
Suppose the three people use plurality voting (each person casts one vote for his
favorite), and, in case of a tie, the social choice is the outcome closest to the beginning of
the alphabet. Under this rule, if 1 votes for his favorite, x, and persons 2 and 3 do
likewise, there is a 3-way tie, which is resolved with the (alphabetical) choice of x. Now
put yourself in the shoes of person 2. You will immediately see that, if persons 1 and 3
continue to vote for their favorites, and if you switch from your favorite y to your second
favorite z, then social choice changes, from x to z, making you better off! You are in
effect being asked “what is your preference relation?” Instead of answering honestly (y z
x), you offer, in effect, a false preference relation (z y x).
Reporting a false preference relation in order to bring about an SCF outcome that
you prefer to the one you get if you are honest, is called strategic behavior, or
strategizing. It is obviously a bad thing if an SCF produces lots of opportunities for
strategic behavior: if individuals are commonly strategizing, there is no reason to
believe that the outcome, based as it is on false reports, is truly best for society. If an
SCF has the property that it is never advantageous for anyone to report a false preference
relation it is called strategy-proof. For instance, suppose an SCF always chooses the
alternative that is first in the alphabetical list of alternatives. This SCF would be
frustrating and idiotic, but it would be strategy-proof.

The Gibbard-Satterthwaite Theorem

This leads to a natural question: Are there SCF’s that are immune to strategic
behavior, and that satisfy a few other reasonable conditions? Note that the question is
very similar in style to the question that Arrow asked about Arrow SWF’s. What would
the reasonable conditions be? First (similar to Arrow), the SCF ought to be universal;
that is, it should work no matter what the profile of individual preferences might be.
Second (also similar to Arrow), there should be no dictator. In the SCF context person i
is a dictator if the social choice is always a top-ranked alternative for person i. Third
(and different from Arrow), the SCF should be non-degenerate. This means that for any
alternative x, there must be some preference profile which would give rise to x’s being
the social choice. (This requirement excludes the SCF that always chooses the first
alternative in the alphabetical list.) Now we can ask the question: Do there exist SCF’s
which are universal, non-degenerate, strategy-proof , and non-dictatorial?
This question was asked and answered, independently, by Gibbard (1973) and
Satterthwaite (1975). The Gibbard-Satterthwaite result turns out to be logically very
close to Third Fundamental Theorem of Welfare Economics; in fact Gibbard uses
Arrow’s theorem to prove his theorem, and Satterthwaite shows that his theorem can be
used to prove Arrow’s. Following is the Gibbard-Satterthwaite theorem. The proof is
omitted; a simplified and restricted version of the theorem, and a simple proof, can be
found in Feldman and Serrano (2006):
Gibbard-Satterthwaite Theorem. There is no social choice function that satisfies
the conditions of universality, non-degeneracy, strategy-proofness, and non-
dictatorship.
Like the Third Fundamental Theorem, the Gibbard-Satterthwaite theorem is starkly
negative; it says that if you want a decision-making process, an SCF to be precise, that
has desirable characteristics, including being immune to strategic manipulation, you are
bound to be disappointed. To put it differently, for any reasonable SCF, there will be
circumstances under which some person will want to falsely report his preferences,
resulting in a perversion of the process, and an outcome that may not be desirable for
society.
If a decision-making process works in a way that offers each individual no
incentive to misrepresent his preferences, no matter what preferences the other n-1
individuals might be reporting, we say that honestly reporting one’s preferences (or
telling the truth) is a dominant strategy. The Gibbard-Satterthwaite result then says that if
a social choice function satisfies the conditions of universality, non-degeneracy and non-
dictatorship, truth-telling will not be a dominant strategy. That is, there will be some
reported preference relations of all individuals except i, which will provide an incentive
for individual i to lie. If everyone else is saying such-and-such (which might be true or
false), person i will give a false report. This is what strategy-proofness excludes.
But what if we narrowed this broad notion of strategy-proofness; what if we
required that i not have an incentive to lie when the others are reporting the truth, rather
than requiring that i never have an incentive to lie, no matter what the others are
reporting?

Implementation and the Maskin Theorem

If telling the truth is a best strategy when others are telling the truth, rather than
always, then truth telling is a Nash strategy, rather than a dominant strategy. The theory
of implementation, or mechanism design, provides a way out the negativity of the
Gibbard-Satterthwaite theorem; it provides a way to implement an SCF, or support its
choices, by incorporating truth telling about preferences into Nash equilibrium strategies
in games.
The major theorem on implementation is due to Maskin (1999), whose paper first
circulated in 1977. In Maskin’s model, there is a social planner (or central authority) who
wants to bring about choices according to a given SCF, which we now call F. The
planner knows F, as do all the members of society. Given any preference profile R, the
SCF produces an outcome F ( R ) = x . But the planner does not know the true preferences
of the individuals. He must rely on the individuals to report their preferences, and they
may lie. We assume for simplicity that every individual knows the true preference
relation for himself and every other individual; that is, each i knows the true preference
profile, but the social planner doesn’t. (This obviously a strong assumption.) From this
point on, when one preference profile may be true and another may be false, we will use
the unadorned R to represent the true profile. The social planner receives reports on
preferences, or preference profiles, from the individuals, but they may be lies. We let Rˆi
represent a reported preference relation for person i, which may be false; similarly
Rˆ = ( Rˆ1 , Rˆ2 ,..., Rˆ n ) represents a reported preference profile, which may be false; and
Rˆ i = ( Rˆ i , Rˆ i ,..., Rˆ i ) represents a preference profile, reported by person i, which may
1 2 n
false. The social planner wants to devise a method, a mechanism, to induce individuals to
honestly report preferences. That way he will get hold of the true preference profile R,
and produce the desired outcome F ( R ) = x .
How might this be done? The intuition is to ask each and every individual to report
a preference profile. (Note that since we assume all the individuals know each other’s
true preferences, it is no more challenging for an individual to report a preference profile,
comprising preference relations for everyone in society, than it is to report his own
preference relation.) If all the reported preferences profiles agree, there’s a good chance
they are all true, and the planner might accept the generally agreed-upon profile. If they
all agree except for one, the one that’s out of line probably comes from a liar, and he
should be given a motive to avoid lying. (If the social planner were a despot, the out of
line person would be shot. Note also that there must be 3 or more individuals in society to
discover whose report is out of line.) Finally, when the reported preference profiles
generally disagree, the social planner needs a way to avoid having the process stop at an
inappropriate Nash equilibrium.
Let us be more precise. Maskin’s algorithm for implementing an SCF works as
follows: Each person i reports a message mi , which is composed of an alternative x, a
preference profile Rˆ i , and a non-negative integer. (i) If every message is the same, of the
form ( x = F ( Rˆ ), Rˆ , 0) , then the social planner chooses x. (ii) If every message but one is
the same, of the form ( x = F ( Rˆ ), Rˆ , 0) for every person but j, while j reports a message of
the form ( y ≠ x , Rˆ j , anything), then the social planner chooses y, unless person j would
like x less than y according to Rˆ j , the person-j preference relation that all the other
people are reporting. If this is the case the planner chooses x. (Person j is not shot. He
simply does not gain, and may lose, from his deviation.) (iii) In all other cases, the social
planner finds the person who proposes the highest integer (with some method for
resolving ties), and chooses the alternative named by that person.
Now the questions can be framed: First, given this mechanism, would
m1 = m2 = ... = mn = ( F ( R), R, 0) , with R the true preference profile, constitute a Nash
equilibrium? Second, if (m1 ,..., mn ) is any Nash equilibrium list of messages in this
mechanism, can we be sure the resulting chosen alternative will be F ( R ) ?
The answers are Yes and Yes, under certain general assumptions. The assumptions
of Maskin’s theorem are as follows: First, there must be 3 or more individuals (so that a
deviant message can be spotted). Second, a mild diversity condition must be satisfied.
Maskin uses a condition called no veto. Loosely speaking, this means that if n − 1 people
prefer x to all the other alternatives, then the SCF must choose x. Alternatively, one can
assume the existence of a private economic good, that everyone values. This guarantees
that individuals will disagree about what alternatives are best. In this entry we will simply
assume diversity, meaning the following: For any given alternative x, there exist at least
two people, each of whom prefers something else to x.
Third, the social choice function F must satisfy an intuitive condition called Maskin
monotonicity. (The condition is actually a distant relative of the NIM assumption used in
the simple version of Arrow’s theorem presented above.) Maskin monotonicity means the
following: Let R̂ and R be any two preference profiles. (These may be true or false; it
does not matter in this context.) Suppose F ( Rˆ ) = x , and suppose that, for all individuals i
and all alternatives y, xRˆ y implies xR y . Then F ( R ) = x . (In other words, in a
i i

hypothetical transition from Rˆi to Ri , for every person i the set of alternatives that i likes
less than x or the same as x has expanded, or at least hasn’t shrunk. Since x was the
social choice under R̂ , x must continue to be the social choice under R .) With these
three conditions, Maskin proved:
Maskin Theorem. Assume n ≥ 3 . Assume diversity and Maskin monotonicity.
Then the mechanism described above implements the SCF F, in the sense that
truthful messages leading to F ( R ) comprise a Nash equilibrium, and in the sense
that any Nash equilibrium list of messages results in the social planner choosing
F ( R) .
We will not provide all of the proof, but the logic is as follows: First, establish
that m1 = m2 = ... = mn = ( F ( R), R, 0) is a Nash equilibrium, where R is the true preference
profile. This is rather obvious, given rules (i) and (ii) of the Maskin algorithm. Second,
establish that under rules (ii) and (iii), there are no Nash equilibria. This follows rather
easily from the diversity assumption. Third, establish that if
m1 = m2 = ... = mn = ( F ( Rˆ ), Rˆ , 0) is any Nash equilibrium, then F ( Rˆ ) = F ( R ) . That is,
given a Nash equilibrium based on a universally reported, but possibly false preference
profile, the outcome implemented is the same as if the true preference profile had been
reported. This follows from the assumption of Maskin monotonicity.
Maskin also provided a near converse this theorem, which says that Maskin
monotonicity is a necessary condition for any SCF F to be implementable. Relatively
simple proofs of both Maskin theorems are available in Feldman and Serrano (2006).
Last Words
Where does welfare economics now stand? The First and Second Theorems are
encouraging results that suggest the market mechanism has great virtue: competitive
equilibrium and Pareto optimality are firmly bound. The Third Theorem exposes
impossibilities and paradoxes in economic choices, voting choices, and, in general,
almost any choices made collectively by society. The Gibbard Satterthwaite theorem, like
the Third Theorem, is a starkly negative result: any plausible social choice function will,
under some circumstances, produce incentives for someone to lie. But the Maskin
theorem is a ray of hope; it suggests a way for a social planner to design a game, whose
Nash equilibria will implement a desired social choice function.

Allan M. Feldman

See also compensation principle; pigou, arthur cecil; public finance; social choice

Bibliography
Arrow, K.J. 1950. A difficulty in the concept of social welfare. Journal of Political
Economy 58, 328-346.
Arrow, K.J. 1951. An extension of the basic theorems of classical welfare economics.
Second Berkeley Symposium on Mathematical Statistics and Probability, ed. J. Neyman,
Berkeley:University of California Press, 507–32.
Arrow, K.J. 1963. Social Choice and Individual Values. 2nd edn, New York: John Wiley
and Sons.
Bergson, A. 1938. A reformulation of certain aspects of welfare economics. Quarterly
Journal of Economics 52, 310–34.
Boadway, R. 1974. The welfare foundations of cost-benefit analysis. Economic Journal
84, 926–39.
Clarke, E.H. 1971. Multipart pricing of public goods. Public Choice 11, 17–33.
Coase, R.H. 1960. The problem of social cost. Journal of Law and Economics 3, 1–44.
Feldman, A.M. and Serrano, R. 2006. Welfare Economics and Social Choice Theory,
2nd edn, New York, Springer.
Gibbard, A. 1973. Manipulation of voting schemes: a general result. Econometrica 41,
587-601.
Groves, T. and Loeb, M. 1975. Incentives and public inputs. Journal of Public
Economics 4, 211–26.
Kaldor, N. 1939. Welfare propositions of economics and interpersonal comparisons of
utility. Economic Journal 49, 549–52.
Kramer, G.H. 1973. On a class of equilibrium conditions for majority rule. Econometrica
41, 285–97.
Lange, O. 1942. The foundations of welfare economics. Econometrica 10, 215–28.
Lange, O. and Taylor, F.M. 1939. On the Economic Theory of Socialism. Minneapolis:
University of Minnesota Press.
Lerner, A.P. 1934. The concept of monopoly and the measurement of monopoly power.
Review of Economic Studies 1, 157–75.
Lerner, A.P. 1944. The Economics of Control. New York: The Macmillan Company.
Lindahl, E. 1919. Just taxation – a positive solution. Translated and reprinted in Classics
in the Theory of Public Finance, ed. R.A. Musgrave and A.T. Peacock, New York:
Macmillan, 1958.
Marshall, A. 1920. Principles of Economics. 8th edn, London: Macmillan, ch. VI.
Maskin, E. 1999. Nash equilibrium and welfare optimality. Review of Economic Studies
66, 23-38.
Mas-Colell, A., Whinston, M.D., and Green, Jerry R. 1995. Microeconomic Theory. New
York: Oxford University Press.
McKelvey, R. 1976. Intransitivities in multidimensional voting models and some
implications for agenda control. Journal of Economic Theory 12, 472–82.
Mises, L. von. 1922. Socialism: An Economic and Social Analysis. 3rd edn, trans.,
Indianapolis: Liberty Classics, 1981.
Pigou, A.C. 1920. The Economics of Welfare. London: Macmillan, Part II.
Plott, C.R. 1967. A notion of equilibrium and its possibility under majority rule.
American Economic Review 57, 787–806.
Samuelson, P.A. 1954. The pure theory of public expenditure. Review of Economics and
Statistics 36, 387–9.
Satterthwaite, M.A. 1975. Strategy-proofness and Arrow’s conditions: existence and
correspondence theorems for voting procedures and social welfare functions. Journal of
Economic Theory 10, 187-217.
Scitovsky, T. 1941. A note on welfare propositions in economics. Review of Economic
Studies 9, 77–88.
Sen, A.K. 1970. Collective Choice and Social Welfare, San Francisco: Holden-Day.